Forward Testing After the Backtest

Why Replay Isn’t the Last Step

Manual backtesting compresses time — its strength for building a sample, and its blind spot for psychology. In replay you know a setup is coming, you can re-watch, there’s no overnight gap of dread, no real money tightening your chest. Forward testing removes the compression: one day of practice per real day, outcome genuinely unknown, execution under true conditions. It’s the bridge between “the strategy has an edge” and “I can trade the strategy.”

What Forward Testing Catches

• Execution friction at real speed — the hesitation on valid entries and the itch to exit winners early that show up as a live-vs-backtest gap.
• Real fills — actual spread, slippage and requotes on your broker, against the backtest’s assumptions.
• Operational reality — alerts, platform quirks, time-zone-correct session timing, the logistics of actually being at the screen when setups fire.
• Your psychology — the variable replay structurally omits, and the one that ends most trading careers.

How to Run It

1. Same rules, frozen. Forward testing validates the backtested strategy, not a new one — no tweaking. Changes mean restarting the clock.
2. Demo or minimal size. A demo isolates strategy/execution; a tiny live account adds real-money psychology at trivial risk. Many traders do demo first, then small-live.
3. 2–4 weeks, or ~20–30 trades — enough to surface execution patterns, not so long that variance dominates the read.
4. Log identically to the backtest so the two are comparable in the same units (net-of-cost expectancy in R). Same journal, same columns.

Reading the Gap

Forward vs backtest	Likely cause	Action
Roughly matches	Strategy + execution both transfer	Scale in gradually
Lower win rate, sizes match	Granularity optimism or fills	Re-validate on ticks; pessimistic fill rules
Trades differ from rules	Discipline gap (you)	More rules practice; fix execution, not the strategy
Everything worse	Backtest had hidden mistakes	Audit the backtest before risking more
Worse but within streak bounds	Normal variance	Keep going — too early to judge

The last row matters most: 20 forward trades can underperform a sound strategy purely by chance. Check the result against the backtest’s worst 20-trade stretch before concluding anything — if you’re inside it, nothing is wrong.

Then What

Match → scale up in steps, re-checking that live stats hold at each size. The progression backtest → forward test → small live → scaled live exists because each stage tests something the previous couldn’t, cheaply, before the expensive version. Skipping forward testing is skipping the only stage that tests you in real time — which is exactly the stage most blown accounts needed.

Frequently Asked Questions

Demo or small live account for forward testing?

Both, in order: demo first to confirm the strategy and execution transfer without real-money noise, then a minimal live account to add the psychology demo can't reproduce. Going straight to small-live is defensible if your risk is genuinely trivial; skipping demo entirely loses a cheap diagnostic stage.

How long should forward testing last?

Long enough for 20-30 trades or 2-4 weeks, whichever gives a readable execution sample for your frequency. Longer mainly adds variance, not insight, at this stage — the goal is catching execution and psychology gaps, not re-measuring the edge you already established in backtesting.

My forward test is worse than my backtest. Do I abandon the strategy?

Not yet — diagnose first. Compare net-of-cost expectancy in the same units, check whether the gap is execution (re-validate fills/granularity), discipline (your trades diverged from the rules), or just variance (inside the backtest's worst-streak bounds). Abandon only after ruling those out over a meaningful sample.

Can I forward test and keep backtesting at the same time?

Yes, on separate strategies — forward test the validated one while backtesting your next candidate. Don't backtest and forward test the same strategy simultaneously, though: discoveries in replay would tempt mid-stream changes to the live test, restarting its clock.